erigontech/erigon

Concurrent Access Issue with freezeblocks.BlockReader leading to Runtime Panic

Closed this issue · 4 comments

SipengXie commented

I encountered a runtime panic when using four threads to concurrently access block hashes with the freezeblocks.BlockReader. The error seems to relate to a CGO pointer issue. Below is the error message that was produced during execution:

panic: runtime error: cgo argument has Go pointer to unpinned Go pointer

goroutine 32 [running]:
github.com/erigontech/mdbx-go/mdbx.(*Cursor).getVal1.func1(0xc0ccfb3d00?, 0xc0ccf02900, {0xc0ccf02900?, 0x28, 0x0?}, 0xf)
	/home/wulab/go/pkg/mod/github.com/erigontech/mdbx-go@v0.27.24/mdbx/cursor.go:210 +0x7a
github.com/erigontech/mdbx-go/mdbx.(*Cursor).getVal1(0xc0ccf46b08?, {0xc0ccf02900?, 0x0?, 0x0?}, 0xc0ccf46b08?)
	/home/wulab/go/pkg/mod/github.com/erigontech/mdbx-go@v0.27.24/mdbx/cursor.go:211 +0x38
github.com/erigontech/mdbx-go/mdbx.(*Cursor).Get(0xc0b7ed5c60, {0xc0ccf02900, 0x28, 0x28}, {0x0?, 0x1ae29b0?, 0x16?}, 0xf)
	/home/wulab/go/pkg/mod/github.com/erigontech/mdbx-go@v0.27.24/mdbx/cursor.go:156 +0x74
github.com/ledgerwatch/erigon-lib/kv/mdbx.(*MdbxCursor).set(...)
	/home/wulab/erigon/erigon-lib/kv/mdbx/kv_mdbx.go:1195
github.com/ledgerwatch/erigon-lib/kv/mdbx.(*MdbxCursor).SeekExact(0xc00007c4e0?, {0xc0ccf02900?, 0x6?, 0xc0ccf46b01?})
	/home/wulab/erigon/erigon-lib/kv/mdbx/kv_mdbx.go:1544 +0x230
github.com/ledgerwatch/erigon-lib/kv/mdbx.(*MdbxTx).GetOne(0x7f72923775b8?, {0x16a8918?, 0xc000100808?}, {0xc0ccf02900, 0x28, 0x28})
	/home/wulab/erigon/erigon-lib/kv/mdbx/kv_mdbx.go:1043 +0x57
github.com/ledgerwatch/erigon/core/rawdb.ReadHeaderRLP({0x72f1536f5458, 0xc00007c4e0}, {0x30, 0xd8, 0x3f, 0x61, 0xc, 0xa3, 0x82, 0xe1, ...}, ...)
	/home/wulab/erigon/core/rawdb/accessors_chain.go:252 +0xc2
github.com/ledgerwatch/erigon/core/rawdb.ReadHeader({0x72f1536f5458?, 0xc00007c4e0?}, {0x30, 0xd8, 0x3f, 0x61, 0xc, 0xa3, 0x82, 0xe1, ...}, ...)
	/home/wulab/erigon/core/rawdb/accessors_chain.go:261 +0x48
github.com/ledgerwatch/erigon/turbo/snapshotsync/freezeblocks.(*BlockReader).Header(0xc0b061a7a0, {0x60?, 0x4?}, {0x72f1536f5458?, 0xc00007c4e0?}, {0x30, 0xd8, 0x3f, 0x61, 0xc, ...}, ...)
	/home/wulab/erigon/turbo/snapshotsync/freezeblocks/block_reader.go:437 +0x79

This issue occurred while trying to read headers using the BlockReader.Header function. Here is the initialization snippet for the BlockReader used in my setup:

func newBlockReader(cfg ethconfig.Config) *freezeblocks.BlockReader {
	var minFrozenBlock uint64

	if frozenLimit := cfg.Sync.FrozenBlockLimit; frozenLimit != 0 {
		if maxSeedable := snapcfg.MaxSeedableSegment(cfg.Genesis.Config.ChainName, SNAPSHOT); maxSeedable > frozenLimit {
			minFrozenBlock = maxSeedable - frozenLimit
		}
	}

	blockSnaps := freezeblocks.NewRoSnapshots(cfg.Snapshot, SNAPSHOT, minFrozenBlock, log.New())
	borSnaps := freezeblocks.NewBorRoSnapshots(cfg.Snapshot, SNAPSHOT, minFrozenBlock, log.New())
	blockSnaps.ReopenFolder()
	borSnaps.ReopenFolder()
	return freezeblocks.NewBlockReader(blockSnaps, borSnaps)
}

Configuration used: cfg = ethconfig.Defaults.

I would appreciate any insights or suggestions on how to address this error. Is there a known issue with concurrent access or a specific setup that I should follow to avoid this panic? Any guidance or updates on resolving this error would be greatly appreciated.

Thank you!

AskAlexSharov commented

can't move tx object between goroutines

SipengXie commented

Can I do dbTx, _ := db.BeginRo(ctx) in different goroutines with a same db?

AskAlexSharov commented

can. must check err

SipengXie commented

Thank u, it helps a lot!